Looping and "Wobble"

When I'm doing a DPBSD analysis on a traditionally looped DP (99% of current offerings) I often think of the schematic of a surf sound generator kit made back in the electronic stone ages by PAIA. In it, a slowly changing "random" voltage controlled a VCA & VCF being fed white noise. The random voltage was produce by summing three non-integer frequency related sine waves, thus producing a somewhat chaotic pattern as our ears have a hard time separating out the relative phases of the fundamentals once three or more sources are summed.

This, to a first order, is somewhat analogous to what is going on with piano notes that have three strings which are slightly detuned. The "wobble" sound is somewhere between chaotic and deterministic. Realistic piano note "wobble" therefore requires a long period so that we can't readily recognize the pattern repeating, but loops are generally only 1 to 4 seconds long. For my ears, and I suspect many others, this is the main reason why most note decays sound fake. But give me a 5 to 10 second loop on the lower notes and I might not notice the period.

Yamaha CP5 Review



I've been begging for samples from the Yamaha CP5 and CP50, mainly to compare the CF voice across these three models, and also to analyze the S6 piano voice found in the CP1 and CP5. PW member "octurn" recently sent me a DPBSD WAV file of the CP5 CF voice, so I went ahead and performed a thorough analysis of it - thanks octurn!

The CP5 passes the pedal down sympathetic resonance test, but the effect sounds somewhat echoy to me. I can particularly hear the echo on the second to last note of the partial pedaling test, where the note goes from partially damped to fully damped. I couldn't detect anything in the way of key-down sympathetic resonance.

It responds to partial pedaling, and has damping samples that play at key and pedal up which sound realistic, though I wasn't able to detect any key / pedal "clunk" or "loom of strings" sound effects in the test file. It passes the brief pedal partial damping test, and passes most steps of the silent replay test, failing only at the end with a note damp at pedal up.

The default setting of the note decay time, which in this DP is adjustable, produces low notes that have a nice long decay, with the mids and highs somewhat shorter. The loop samples are all quite short, not long enough to convincingly "wobble" or to suppress the audibility of the loops. I had a very difficult time determining the loop lengths for the upper mids and highs, suggesting very short lengths, heavy loop pre-processing, or something else entirely.

This sample set is fairly stretched, with 48 stretch groups covering 88 notes. The stretching is restricted to the low and high ends, with the mids unstretched. I can hear the lower stretch group transitions, though they don't have as much timbre variation between them as many other DPs, which is good. I think I would have preferred groups of 2 all the way up, but that probably wouldn't have reduced the sample set size as much.

In terms of velocity layers, this appears to be a blended 4 or 5 layer sample set, with only the highest layer transition fairly audible to me.

It's clear from the above, and also from the intentionally lighter and ungraded key action, that the CP1 and CP5 are targeted more towards stage musicians who need a variety AP, EP, and organ sounds, rather than pure AP enthusiasts. It should be mentioned that there are various interesting parameters that can be adjusted in the AP voices, such as hammer hardness. While this is highly welcome and certainly adds value, the support of these adjustments may have forced cuts in the fundamental sample set size.

MP3 and all analysis pix here:
mp3: http://www.mediafire.com/?l8qpam2oeeoi2q9
pix: http://www.mediafire.com/?odlhzoufh2v9rrp

Some analysis pics and the text review follows. Many thanks again to octurn for the DPBSD file!


Spectral phase view of the pedal down sympathetic resonance test, stimulus removed. On the left the pedal is down, on the right it is up. The effect sounds somewhat echoy to me.


Waveform view of the looping test with vertical zoom applied. Lower decay times are nice and long, mid and highs are somewhat short.


Spectral pan view of the looping test, note C2. Attack and loop samples are clearly seen, cursor is located at the transition point, looping is fairly audible in the low and mid notes.


Spectral pan view of the stretch test, low notes. A fair amount of stretching going on, audible in the lows.


Spectral frequency view of layer test. Evidence of 4 or 5 layers, with only the highest layer transition audible.


--------------
- Yamaha CP5 -
--------------
FILE & SETUP:
- dpbsd_v1.8_yamaha_cp5_cf.mp3
- Default patch 1 CF Grand with MOD-FX block & reverb disabled.
- Sequenced & recorded directly using CP5 & thumbdrive, WAV=>MP3 via Audition.
- Recorded by "octurn".
PROS:
- Passes the pedal down sympathetic resonance test.
- Passes the brief pedal partial damping test.
- Responds to partial pedaling.
- Key/pedal up note damp samples.
- Lowest notes have long decay.
- This is a blended 4 or 5 layer sample set.
- Visible layer switch @ vel=60,100,122.
CONS:
- Pedal down sympathetic resonance sounds echoy.
- Fails the key down sympathetic resonance test.
- Fails the pedal down silent replay test @ pedal up (note damps).
- Mid and high notes have somewhat short decay.
- Obviously looped, both visibly and audibly.
- Attack and loop sample lengths are fairly short.
- Attack sample lengths are (C1:C8): 2(?),2.1,2.1,1.8,1.7,1.3,0.8,? seconds.
- Loop sample lengths are (C1:C8): 0.8,0.8,0.6,0.5,0.4,?.?,? seconds.
- Audibly stretched over the low end.
- Stretch distances: 3,3,4,4,4,2,4,3,1(x26),2,3,2,1,1,1,3,2,2,3,4,2,3,3,3 = 49 groups.
- Audible layer switch at vel=122.
- No obvious pedal up/down "loom of strings" or key up "clunk" sounds.
OTHER:
- Notes played @ vel=1 produce a sound.
- Dynamic range 44.3dB (vel=1:127).
- MP3 levels: peak @ -1 dB, noise floor below -80 dB.
- Date reviewed: 2010-09-20, revised 2010-10-02 (layer switch points).


[edit]Fixed velocity layer switch points.

Interesting stuff. Thanks to yourself for the analysis and to octurn for the sample.

So, how does this compare with the CP1?

Cheers,
James
x

Good question. Same stretch group pattern, same sample lengths, same layer count and transition points. I need to spend some time comparing them a bit more, but at this point they seem identical.

[edit]Oops, two of the velocity layer switch points are slightly different between the CP1 & CP5. Sorry about the confusion everyone, mistakes were made.

Well, that's not so surprising. I expect the extra money goes on those additional effects processors.

Cheers,
James
x

I agree...interesting stuff. The subjective difference noted by some between the CP1 and CP5 must be due to signal path differences such as the amp blocks etc...not that this should have any relevance at all to acoustic pianos patches but a few people have reported what they perceived as differences...or maybe it was just down to the other variable parameters such as hammer hardness etc.

One final point Dewster...when you refer to adjustable decay it is my belief that the adjustable bit is indeed the tail of the note - but not when sustained. I don't believe the length of the sustain is adjustable. So a note struck and then released has a decay period, albeit very short...it is this portion that I believe is adjustable. Attack and decay times are adjustable on the RD...or at least they were on the old RD-600. The Nord could use longer decay on the Yamaha Studio Grand 2...a note quickly struck then immediately released cuts off very sharply indeed...almost like an electronic cut-off. I could be wrong about the CP of course!

Cheers,

Steve

@EssBrace

Both of the parameters you mentioned are adjustable on the CP5. Those parameters are called "decay time" and "release time".

Decay Time can be used to adjust how fast a note decays while the key is being held down.

Release Time is used to adjust how fast a note decays after the key is released.


Furthermore, I'm not sure which amp blocks would be on the CP1 and not on the CP5?

Ok, thanks for clarifying...so you could adjust the piano to sustain indefinitely I guess...which presumably means it just keeps playing the looped part...

I've got it into my head that there is additional pre-amp stages on the CP1...thought I'd read that somewhere...again, although much is made of this by Yamaha I don't see an acoustic piano application for that but I suppose the AP patches can be routed this way?

Steve

I agree, Steve. This spoils an otherwise very good sample. To me, it makes playing the Nord slightly more "synthy" than I would like - the illusion of the piano is somewhat lost. However, because Nord updates their samples and OS, I know that things will likely get better over time - how many DPs can you say that about, including mega expensive ones like the CP1?

A Tale of Two CPs: Yamaha CP1 vs CP5

Now that we have a DPBSD MP3 for the CF voice in both the Yamaha CP1 and CP5, we can compare them to see how similar this voice is in these two DPs. I noted in an earlier post that the sample lengths, stretch groups, and layer switch points are the same between them. This is true except for the layer switch points - I did a more careful analysis of both today and found slight differences between the two (sorry about that misinformation):

CP1: Visible layer switch @ vel=54,78?,98,122.
CP5: Visible layer switch @ vel=60,100,122.

And now for a more visual comparison, I took some analysis pix with the same horizontal (time) scale, and with the amplitude processed in the same manner.



Spectral pan view of the stretch test, the unstretched middle note range, CP1 at top, CP5 at bottom. Amplitude normalized to -1dB to bring out detail. No significant visible or audible differences.



Spectral pan view of the layer test, CP1 at top, CP5 at bottom. Amplitude compressed 20:1 to bring out detail. Other than a small difference in two layer swich points, there are no significant visible or audible differences.



Spectral phase view of the layer test, CP1 at top, CP5 at bottom. Amplitude compressed 20:1 to bring out detail. Other than a small difference in two layer swich points, there are no significant visible or audible differences.

Next, I took the note C2 from both MP3 files, collapsed them to mono, normalized them to -1 dB, and matched the starting points in time as closely as I could:


Waveform view of the start of note C2, CP1 at top (left channel), CP5 at bottom (right channel). No real difference in the time domain at the start of the attack sample, which is kind of amazing after all the compression, conversion, etc. the samples have been through.

As the notes decay, the samples aren't exactly the same, but the difference appears to be largely one of a slight sampling rate difference, which isn't too unexpected given the very different recording setups:


Spectral phase view of the note C2, CP1 in the left channel & CP5 in the right. Monotonic phase change shows simple difference in sampling rates between the two samples, one phase period is ~3.8 seconds.


Waveform view of the note C2 @ 3.8 seconds in, CP1 at top (left channel), CP5 at bottom (right channel). Period difference is visible, with the CP1 delayed one cycle of the fundamental frequency, but only 1/2 cycle of the detailed cycle, waveforms are essentially identical.


Waveform view of the note C2 @ 7.9 seconds in, CP1 at top (left channel), CP5 at bottom (right channel). Period difference is visible, with the CP1 delayed one cycle of the detailed cycle. Even this far into the decay the waveforms are essentially identical.

So, if you buy a CP5 are you getting the same CF sample set as in the CP1? I believe the answer to this is yes, though the slight difference in two of the layer switch points confounds things a bit.

Anyone got a CP50 sample to offer up?

Interesting material ;-)

It is not possible that the shift could be caused by a slight difference in the CP1 vs. CP5 settings concerning the attack time and/or decay time?

Well, at 7.6 seconds the phase offset is 0.015 seconds:

1,000,000 * ( 0.015 / 7.6 ) = ~2000 ppm.

That's more than I would expect from two different crystal controlled timing references (typically 50 ppm for consumer grade stuff). I'm not sure what is causing that much phase difference, it's almost like one of the DPs was slightly detuned.

Any interest in old keyboards? I have and old Korg (X5D) hanging around, I'd guess analyzing that is a complete loss of time but maybe you'd like the database to be really comprehensive.

I'd be happy to analyze it!

It's actually kind of scary how little things have changed over the decades, particularly sample memory size.

Let things be scary, then.

I have a venerable Kurzweil Micropiano MIDI module. I bought it second hand, but I think it was released in 1992! 18 year ago. And guess what, for the age it has, it has stood quite well the pass of time.

I will make a DPSBD MP3 of it one if this days then...

Your honor, I rest my case. If we continue at this rate, at some point in the future we'll be unearthing fossilized DPs that compare favorably with contemporary fare.

Yamaha CP5 - BONUS!

A quick analysis of the S6 voice in the Yamaha CP5, many thanks to octurn for this second DPBSD file! Ever since Yamaha put this voice in the CP1 & CP5 I've been rather anxious to hear it up close, so it's great to finally have a sample of it.

MP3 and all analysis pix here:
mp3: http://www.mediafire.com/?w3dyi28vq6his9s
pix: http://www.mediafire.com/?6i1r7qc1pdxo11s


Figure 1. Spectral frequency view of the looping test, notes C2 through C6.

I want to talk about this first because it is quite strange. Several seconds into the decay of these notes, it sounds like a low pass filter is abruptly imposed on the playback. After spending some time listening to the looping test and watching the VU meters in Audition, it is clear to me that this is due to poor timbre matching between the attack and loop samples, because it happens right at the point where the the attack sample has fully crossfaded over to the pure loop sample. The cursor is placed on C2 where the timbre change happens, and you can easily see - particularly in the black & white view above - and hear this point in the other notes as well.


Figure 2. Spectral pan view of the pedal down sympathetic resonance test, stimulus removed. On the left the pedal is down, on the right it is up. The effect is a bit echoy, but it does help smooth out the sound of the attack / loop timbre anomaly discussed above.


Figure 3. Waveform view of the looping test with vertical zoom applied. Decay times are nice and long.


Figure 4. Spectral phase view of the looping test, note C3. Attack and loop samples are clearly seen, cursor is located at the transition point. Loops are processed and rather bland, looping is fairly audible in the low and mid notes.


Figure 5. Spectral pan view of the stretch test, mid notes. A fair amount of stretching going on across the entire range, audible in the lows.


Figure 6. Spectral pan view of layer test, compressed 20:1. Evidence of 4 layers, with two transitions somewhat audible.


Figure 7. Spectral frequency view of layer test, compressed 20:1. The layers are not completely blended, so timber change with velocity is not as smooth as other Yamaha piano sample sets.


FILE & SETUP:
- dpbsd_v1.8_yamaha_cp5_s6.mp3
- Default patch 1 S6 Grand with all mod & amp blocks & reverb disabled.
- Sequenced & recorded directly using CP5 & thumbdrive, WAV=>MP3 via Audition.
- Recorded by "octurn".
PROS:
- Passes the pedal down sympathetic resonance test.
- Passes the brief pedal partial damping test.
- Responds to partial pedaling.
- Key up "clunk" sounds.
- Key/pedal up note damp samples.
- Notes have nice long decay.
- I believe this is a somewhat blended 4 layer sample set.
- Visible layer switch @ vel=70,98,120.
CONS:
- Pedal down sympathetic resonance sounds echoy.
- Fails the key down sympathetic resonance test.
- Fails the pedal down silent replay test @ pedal up (note damps).
- No obvious pedal up/down "loom of strings" or "clunk" sounds.
- Obviously looped, both visibly and audibly.
- Loop sample lengths are fairly short.
- Notes C2, C3, C5, C6 have poor timbre match between attack and loop samples.
- Attack sample lengths are (C1:C8): 3.7,3.7,3.2,2.8,2.4,2.1,0.6,? seconds.
- Loop sample lengths are (C1:C8): 0.8,0.7,0.6,0.6,?,?,0.8,? seconds.
- Audibly stretched over the low end.
- Stretch distances: 2(x8),3,2(x3),3,2,2,1,2(x3),1,1,2(x3),1,2(x20) = 45 groups.
- Somewhat audible layer switch at vel=70,98.
OTHER:
- Notes played @ vel=1 produce a sound.
- Dynamic range 42.5dB (vel=1:127).
- MP3 levels: peak @ -1 dB, noise floor below -80 dB.
- Date reviewed: 2010-10-08.

Dewster:

Pianoteq 3.6.5 has been released. I've rendered your midi file to wave/mp3 for two pianos - the C3 and the M3:

As requested, no reverb, and files are near max, both are Layer 3 ACM, 44100 Hz, 192 kbps, stereo.

http://www.box.net/shared/jpjn0jx5kx

http://www.box.net/shared/4hymy5xy7z

Glenn

Changelog for geeks (myself included):

Version 3.6.5 (2010/10/20)
- Key range of K1, C3 and M3 grand piano extended to 105 keys.
- Keyboard transposition can be controlled with the computer keyboard, or custom MIDI mappings.
- Pianoteq Standalone now recalls both A and B presets on startup.

I renamed them to include the Pianoteq version and placed them in the DPBSD MP3 archive.

I listened to them this morning. Like the K1 I tested back in March, both pass all my tests. Other than that there isn't a lot to say, other than the M3 is quite bright and has a slight wah-wah pedal or phasey sound when notes are played together, and the C5 decay sounds overly muffled to me. Both have a strange periodic buzzy sound when partially pedaled.

Thanks Glenn!

Dewster:

I just noticed that you have not done an analysis on the Roland KR7 (which I have owned for a few years). I believe it is very similar to the RF7.

Would you like a mp3 of the KR7?

Glenn

PS - on the Pianoteq mp3: The K1 I sent you utilized an unusual mic position (only two mics). The "official" setup uses four mics. Perhaps I could also do another K1 using the "official" mic setup?

Yes to both if it isn't too much trouble. Thanks!

I definitely think the APs on the CP5 and the CP1 are the same. You can also see the signal path chart for APs on the CP1 is identical to the one on the CP5 per Yamaha's website. On the other hand the EPs are definitely different because of the extra signal processing available. Nevertheless, the underlying sample set is probably the same. I don't really know for sure.

Dewster, how difficult would it be to include an "effective polyphony" metric in the test for the various options available on the boards? For example,

Rated Polyphony x
Polyphony with Stereo
"" and layered stereo sound
"" and string resonance
"" and extra long decay
"" etc. etc.

Good idea, but I think that would be a very difficult thing to analyze. As you know, note stealing algorithms by their very nature try their best to hide what they are doing from the player.

Playing lots of notes together would also increase the peak recording level, which would bury the decay tails even deeper into the noise floor. I guess it could be a separate file, but I've been doing my best to avoid that.

A Tale Of Two Notes: Real vs. SN vs. UPHI vs. layered UPHI
(edited to include layered UPHI)

This is a quick comparison of two notes (C3 and C4) rendered four different ways:
1. VintAudio C7 Close Miked 6 layer voice - what I'm calling the "real" piano in this group.
2. Roland RD-700GXF SN01 Grand Piano voice - the SuperNATURAL or SN piano in this group.
3. Kawai CA63 Concert Grand voice - the Ultra Progressive Harmonic Imaging or UPHI piano in this group.
4. Kawai CA63 TADutchman's BLENDED CONCERT GRANDEUR patch - the layered UPHI piano in this group.


- Real vs. SN -


Figure 1. Spectral pan view of 6 seconds each of (left to right) real C3, SN C3, real C4, SN C4. Note groups have been normalized to -1dB but are otherwise untouched.

I invite you to grab some good headphones and give a listen to the MP3 associated with this view, which is located here:
mp3: http://www.mediafire.com/?m5blnga8mqb48iz

Things I hear:
1. The "phasey" sound of the real decay is also present in the SN decay.
2. The timbre or brightness of the decay in both doesn't abruptly change, but evolves over time.
3. There is some smearing in the SN attack, as if the beginnings of the notes go on a bit too long or something.
4. The sound of the key action / hammer / strings is clearer in the real piano.
5. This particular SN piano is obviously a "darker" sounding piano than a Yamaha C7.
6. I prefer the sound of the real piano.


- Real vs. UPHI -


Figure 2. Spectral pan view of 6 seconds each of (left to right) real C3, UPHI C3, real C4, UPHI C4. Note groups have been normalized to -1dB but are otherwise untouched.

The MP3 associated with this view is located here:
mp3: http://www.mediafire.com/?w41n92904rpro8d

Things I hear:
1. The "phasey" sound of the real decay is not present in the UPHI decay.
2. There is a noticeable timber change as UPHI transitions from the attack to the loop phase of the note.
3. The UPHI decay itself is somewhat quick - the timbre sounds muffled and static.
4. Looping is clearly audible in UPHI note C4.
5. I really prefer the sound of the real piano.


- Real vs. layered UPHI -


Figure 3. Spectral pan view of 6 seconds each of (left to right) real C3, layered UPHI C3, real C4, layered UPHI C4. Note groups have been normalized to -1dB but are otherwise untouched.

The MP3 associated with this view is located here:
mp3: http://www.mediafire.com/?givayaajxohk5xv

Things I hear:
1. The "phasey" sound of the real decay is not present in the layered UPHI decay. What is there instead is an unfortunate emphasis of the loop itself - which is too short - and so sounds more like old school looping to me, particularly on layered UPHI note C3.
2. I'm not a fan of either static or pronounced loops, but if I had to pick one I think I would go with the unlayered UPHI sound over the layered UPHI sound.
3. I really prefer the sound of the real piano.

I agree that a renowned grand piano (of almost any brand) beats a DP, interesting comparison though.
The biggest limitation here is that you are subjectively comparing timbres/sound signatures of completely different brands. (I know, Roland doesn't do AP's).

As far as console type DP's is concerned, you have obviously used the single voice mode factory standard grand piano preset of both the Roland and the Kawai. Only the Kawai CA93/CA63 offer a dynamic dual voice mode in combination with UPHI, yielding an enormous versatility in piano timbres and other dynamic sounds currently not possible with any other console type DP (and way beyond the standard factory presets). Reference: http://www.attacca.eu/images/settings_CA63_93.pdf

Assuming you can handle it in a fair way, I could deliver a few C3 and C4 notes (6sec. WAV format) in CA93/CA63 dual voice mode.
Just let me know your thoughts.

I think that sounded pretty snarky.

It would be interesting indeed to see if layering two looped piano sounds would start showing the phasey kind of decay we seen on an acoustic piano and SN. I for one would like to see this.

Left or right one?

The main piano voices briefly compared here are single instances (i.e. not layered) for two reasons:
1. The expectation is that DP manufacturers put their best foot forward on their first patch, so it is a fair candidate for comparison.
2. A single instance lets everyone hear what the base samples sound like, unobscured by phase effects, etc.

If layering two pianos brings some life to the relatively (when compared to the real thing) dead loops of the CA63/93 that's great, but it wouldn't be a level playing field for comparison. Also, I'm not sure I would want to layer the VintAudio or SN pianos as they sound fairly realistic to me as-is.

That said, if you want to provide WAV or MP3 samples (with reverb off) I would consider adding them to the post, with the above caveats noted.